Speech Pattern Recognition Models for Unpunctuated Question Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech recognition systems struggle to efficiently identify questions in unpunctuated transcribed speech files, leading to inefficiencies in processing and understanding conversational data.

Innovation Solution

The system employs a machine learning model that processes transcribed speech by identifying utterance-delimited multiword units, generating utterance-based data vectors, and determining question-indicative word patterns to accurately identify questions within the speech.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional speech recognition systems process unpunctuated transcribed speech files, then the system can handle raw speech data without punctuation, but the system fails to efficiently identify questions within the speech

Engineering Contradiction:

Improvequestion identification accuracyVSAvoidprocessing efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system segments the unpunctuated speech text into utterance-delimited multiword units based on pause detection, creating manageable segments that can be individually analyzed. This segmentation allows the machine learning model to process smaller, structured units rather than attempting to analyze the entire unpunctuated text stream, thereby improving both accuracy and efficiency.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system performs preliminary processing by transcribing speech to text and delimiting utterances before question identification. This preliminary action of structuring the data with utterance delimiters based on speech pauses prepares the data in advance, enabling the machine learning model to focus specifically on identifying question patterns without having to simultaneously handle transcription and punctuation issues.

Inventive Principle:

Principle #10Preliminary action

2Measurement precision

If the system uses machine learning models to identify question patterns, then question identification accuracy improves, but computational complexity increases

Engineering Contradiction:

Improvequestion identification accuracyVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The system divides the complex task of question identification into manageable segments by processing utterance-delimited multiword units individually. This segmentation reduces the computational burden on the machine learning model, as it only needs to analyze smaller, structured units rather than entire paragraphs of unpunctuated text, thereby managing complexity while maintaining accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system introduces an intermediary structure of utterance delimiters and multiword units between the raw speech input and the machine learning model. This intermediary layer preprocesses and structures the data, making it more suitable for machine learning analysis and reducing the complexity of patterns the model must directly interpret from raw unpunctuated text.

Inventive Principle:

Principle #24Intermediary (Mediator)

3Adaptability or versatility

If the system processes speech without punctuation marks, then it can work with raw transcribed data, but it loses structural information needed for accurate question identification

Engineering Contradiction:

Improvehandling of unpunctuated speechVSAvoidstructural information

Core Design Contradiction:

Adaptability or versatilityVSLoss of information

Solution Approach 1:

The system performs preliminary delimitation of utterances based on speech pauses before question identification. This preliminary action preserves structural information by marking where one utterance ends and another begins, even in the absence of punctuation. The utterance delimiters serve as structural markers that maintain the organizational framework needed for accurate question identification while still working with unpunctuated transcribed data.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system introduces utterance delimiters as an intermediary structure between the unpunctuated transcribed speech and the question identification process. These delimiters serve as structural markers that preserve information about speech boundaries and utterance organization without requiring traditional punctuation, thereby maintaining adaptability to unpunctuated input while preventing loss of structural information.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12347426B2Computer-based systems utilizing speech pattern recognition machine learning models and methods of use thereof

Publication Date: 2025.07.01 CAPITAL ONE SERVICES LLC

AI summary

A method receiving at least one transcribed speech file of at least one person; identifying a plurality of utterance-delimited multiword units in the at least one transcribed speech file; generating a plurality of utterance-based data vectors corresponding to the plurality of utterance-delimited multiword units by assigning a numerical data token to each word in each utterance-delimited multiword unit; determining at least one utterance-delimited multiword unit from the plurality of utterance-delimited multiword units having question-indicative word patterns by inputting the plurality of utterance-based data vectors into at least one machine learning model, configured to identify the question-indicative word patterns by computing a probability of a word or groups of words in each of the plurality of utterance-delimited multiword units; and identifying at least one question in the at least one transcribed speech file of the speech of the at least one person using the question-indicative word patterns.